Comparison
Data downtimevsSilent failure
Data downtime
the numbers were wrong from Tuesday 02:00 until someone noticed on Thursday afternoon, and that gap is the thing worth measuring.
The period during which data was missing, wrong or stale, measured from when it broke to when it was fixed. Naming it as downtime is the useful move: it makes detection time visible as a separate number from repair time, and detection time in data teams is routinely measured in days. Almost every investment in quality checks is an argument about reducing the first number.
Full entry →Silent failure
every task is green, the run took its usual eleven minutes, and it processed zero rows.
A pipeline reporting success while producing nothing or producing nonsense. It is the characteristic data failure, because most pipeline code cannot tell the difference between 'no new data' and 'the filter is broken'. The only defences are assertions about the output rather than about the run — freshness, volume, distribution — which is why quality checks are operational monitoring in this domain rather than a testing nicety.
Full entry →